How does Curtly help on classification tasks?

By pruning prompt boilerplate and constraining outputs to direct labels, cutting token counts by 50% or more.

Can I route high-volume webhook queues through Curtly?

Yes. Curtly is built for high-throughput concurrency and continuous streaming.

OpenAI · GPT-5.6 Luna · o200k_base

GPT-5.6 Luna token cost optimizer and calculator

Extreme efficiency for high-frequency micro-tasks, classification, and embeddings preprocessing.

Input token price

$0.20/1M

Output token price

$1.20/1M (6x)

Tokenizer family

o200k_base

Average combined savings

Up to 65%

Why GPT-5.6 Luna generates high API invoices

GPT-5.6 Luna is OpenAI ultra-fast micro-model ($0.20/1M in, $1.20/1M out). In extreme throughput pipelines processing tens of millions of records daily, even fractions of a cent per request total thousands of dollars in serverless bills.

Massive scale multiplier

Pipelines processing 50M+ requests per month multiply tiny token inefficiencies into substantial cloud bills.

6x output generation ratio

Output tokens cost $1.20/1M. Verbose formatting on simple classification tasks unnecessarily doubles compute expenses.

Repetitive categorization templates

Prompt templates with static examples repeatedly send redundant classification rubrics.

o200k_base high-throughput profile

Efficiency rating: 200k vocabulary, ultra-fast token processing

Modern 200k vocabulary. Extremely efficient when input prompts are stripped of redundant template framing.

How Curtly optimizes GPT-5.6 Luna workflows

Template boilerplate elimination

Prunes redundant classification instructions and static few-shot examples without hurting accuracy.

Concise output enforcement

Constrains Luna to return direct classifications or JSON without conversational framing.

In-memory streaming speed

Sub-2ms execution matches Luna extreme inference speed.

1-line drop-in integration (typescript)

import OpenAI from 'openai'; const client = new OpenAI({ apiKey: process.env.OPENAI_API_KEY, baseURL: 'https://proxy.curtly.dev/v1', defaultHeaders: { 'X-Curtly-Key': process.env.CURTLY_API_KEY }, }); const completion = await client.chat.completions.create({ model: 'gpt-5-6-luna', messages: [{ role: 'user', content: 'Classify sentiment: Great delivery speed' }], });

GPT-5.6 Luna ROI simulator